I can predict cars CRASHING. And it's 99% accurate!

I built a machine learning model that predicted with 99.99% accuracy whether a person would die in a car crash on any given day.
Or did I?
Out of hundreds of metrics, here are 5 of the most important ones you'll ever use in machine learning.
🔔 Subscribe for more stories: www.youtube.com/@underfitted?...
📚 My 3 favorite Machine Learning books:
• Deep Learning With Python, Second Edition - amzn.to/3xA3bVI
• Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow - amzn.to/3BOX3LP
• Machine Learning with PyTorch and Scikit-Learn - amzn.to/3f7dAC8
Twitter: / svpino
Disclaimer: Some of the links included in this description are affiliate links where I'll earn a small commission if you purchase something. There's no cost to you.

[@dimasveliz6745]

Nice dynamics man, awesome the seconds where you stopped so we can cope with your explanation or pause the video.

[@underfitted]

Thanks, Dimas!

[@PritishMishra]

Very creative way to show imbalanced dataset problem. The video is very informative as well.

[@underfitted]

Glad it was helpful!

[@edmundfreeman7203]

Accuracy, precision, recall and F1 all depend on the positive/negative boundary. That can be highly context dependent. What are the costs and benefits of the different options? This may or may not be the most likely prediction. I.E. the model may want to label something as positive at a 51% prediction but the cost/benefit tradeoff may be at 20%.

[@ashwinikumar2492]

Loved this video! Have an interview tomorrow, and these examples are going stay through.

[@underfitted]

Good luck in your interview!

[@dexkconsultoria-solucoesem7296]

This content is amazing, CONGRATS

[@Param3021]

hmm, looks like today's and tomorrow's questions are related to this video

[@underfitted]

They are! It wasn't on purpose.

[@miguelhinojosa5594]

Awesome style!

[@underfitted]

Thanks, Miguel!

[@DanielTobi00]

Thank you for sharing

[@underfitted]

You bet!

[@ashwinshetgaonkar6329]

Very informative

[@underfitted]

Glad it was helpful!

[@Param3021]

Thanks for revising me these concepts! I have a question (I am beginner, so the question is maybe dumb) I am working on Heart Disease prediction problem, and I know it's a rare case problem, so data maybe imbalanced. But I have balanced training dataset, it's ratio is 0.51:0.49 (class 1 : class 0). I have used neural networks to make the model and it's accuracy is 98% on validation data. So, I am confused, should I believe the accuracy or use F1-score as a main metric. I am confused because as it's rare event, but in the data, it's not rare, what should I supposed to do then ?

[@underfitted]

In your case, this problem is not imbalanced because your dataset is properly balanced. Accuracy should work for this dataset. But, in real life, heart disease is rare, so you'll find that your model will mostly see negative samples. Therefore, I'd recommend you focus on the F1-score.

[@Param3021]

@@underfitted ohk thanks! Now I get clarity, I will use F1-score as a main metric. Thank you so much sir 🙏

[@edmundfreeman7203]

Can you simulate the effect of actually using your model in your application?

[@talkingbirb2808]

wow that was an unexpected turn

[@underfitted]

Was it?

[@DanielTobi00]

At what percentage would you say a dataset is imbalance.

[@DanielTobi00]

I am referring to the target variables

[@underfitted]

Hard to specify a threshold, but usually we refer to problems where there's a significant difference between classes. For example, 10,000 samples of class A and 11,230 samples of class B is not an imbalanced problem. But 10,000 class A samples and 10,000,000 of class B is.

[@DanielTobi00]

@@underfitted thank you.

[@DanielTobi00]

@@underfitted thank you.

[@anthonylovesmary]

Such strange video and audio editing choices. I'm assuming that this is done for a youtube engagement related reason, but i find it so distracting. The constant zooming in and out of his face, turning his head upside down??, never staying on any graphic long enough for viewer to absorb content, weird background noises punctuating words. Maybe this is how younger audience brains have been trained to absorb information. Very curious about the benefits of this disorientation-based content creation trend.

[@underfitted]

Thanks for taking the time and writing your feedback! I really appreciate it!